How accurate is a DEXA scan for body fat?

How accurate a DEXA scan is for body fat depends on what you hold it up against, and no single method has been agreed on as the benchmark. A position statement for the International Olympic Committee's medical commission reviewed a century of body composition techniques and concluded that none of them is a universally applicable criterion or gold standard, because every method either measures fat indirectly or leans on assumptions about the body that are not true for everyone [1].

In practice the scan does that job anyway. Reviews of body composition methods describe dual-energy X-ray absorptiometry as the reference technique of clinical practice, the one cheaper methods get validated against, and a whole-body scan sorts you into three parts: bone mineral, fat mass and fat-free mass [5].

Measured against one shared yardstick, the methods spread out more than a clinic printout implies. A meta-analysis of 54 studies drawn from work published in 1985-96 compared body fat readings with underwater weighing, which works fat out from how buoyant a body is, in healthy white adults [14]:

• a four-part reference model read 0.6 percentage points higher
• a three-part model read 1.4 points higher
• heavy-water dilution, which infers fat from how much water the body holds, read 1.5 points higher
• one manufacturer's DEXA scanner read 7.2 points higher, with the gap running from roughly 3 to 12 points across studies

That 7.2-point figure belongs to one brand of 1990s hardware, not to scanners sitting in clinics now. It does illustrate the analysis's broader conclusion: the method and the machine move the number, and some methods tilt consistently in one direction [14]. Bioimpedance equations in the same analysis sat 2.0 points above underwater weighing and 2.6 points below it depending on which equation was used, and skinfold measurements ranged from no difference at all to 1.2 points low [14].

Against the standing bioimpedance scales sold as the cheap alternative, which pass a small current through the body and infer fat from the resistance, the scan reads higher. Among 711 adults in Taiwan, one standing analyser put average body fat at 22.5% while the scan put it at 26.3%, so the scale read about 3.7 percentage points lower; it under-read in every group except lean men, where it ran 0.45 points high [6]. Three bioimpedance models tested on 67 healthy adults over two visits one to three days apart all leaned the same way, under-reading fat and over-reading fat-free mass compared with the scan [2].

Repeatability is a different question, and that study answers it for the cheaper devices rather than for the scan: a person had to shift by roughly 2 to 3 percentage points of body fat before one of those machines could call the change real [2]. For separating fat under the skin from fat packed around the organs, and muscle from the fat inside it, a review of body composition methods found MRI, which images the body using magnetic fields, and CT, which does it with X-ray slices, accurate enough to tell those compartments apart, at high cost and with no portability [3]. The cheapest route is a tape measure: an equation using only height and waist circumference was validated against DEXA-measured body fat in 3,456 American adults and estimated body fat better than body mass index, which is weight divided by height squared, missing fewer cases of obesity [4].

The disagreement is built into how each method works. A DEXA scan sends two X-ray beams of different energies through the body and measures how much of each is absorbed; fat, bone mineral and lean tissue absorb them in different ratios, so the machine can solve for three quantities at once [5]. It never weighs fat directly, it infers it from that absorption pattern.

Underwater weighing, heavy-water dilution and bioimpedance each infer fat from something else again: how dense the body is, how much water it holds, how easily it conducts electricity. Every one of them needs a fixed assumption about the fat-free part of the body, usually its density or its water content, and that assumption is not the same in every person. The Olympic committee statement put the condition plainly: a multi-part model only works as a criterion if its calculation accounts for how much the density of fat-free tissue varies between people [1].

This is why a device can be consistent and wrong at once. The three bioimpedance analysers tested against DEXA reproduced their own answers closely, yet all of them sat below the scan on fat and above it on fat-free mass, and the size of that error changed with body size: for fat mass in women, and for fat-free mass in men [2]. Bioimpedance also loses accuracy when the equation built into the device does not suit the person being measured, or when the reading is taken outside a standard protocol [5].

The method-comparison evidence is narrower than the question people put to it.

• The 54-study meta-analysis covered healthy white adults and drew on work published in 1985-96, so its DEXA number describes one manufacturer's scanner of that era [14].
• The largest bioimpedance comparison ran entirely in Chinese participants in Taiwan on a single standing analyser [6], and the three-device study had 31 men and 36 women, all healthy [2].
• None of these papers reports how far apart two DEXA scans of the same person land, so the scan's own repeat-measurement error is not something this evidence settles.
• The comparison work is adult. A systematic review in 7-to-10-year-olds found that simple measures such as body mass index and waist circumference correlate only moderately with a DEXA body fat reading, so adult figures do not transfer to children [13].

Two limits attach to the scan rather than to the research. Reviews of the method list cost and the need for trained staff as practical constraints, and note that no more than two whole-body scans a year were advised [5]. And none of these papers tested whether knowing a body fat percentage changes anyone's health; they compared measurements against each other.

Longevity clinics and body composition studios sell the whole-body DEXA scan as a standalone service, and the clinics directory on this site shows which ones list it. A first appointment is brief. You change into clothing with no metal on it, lie still on a padded table for a few minutes while an arm tracks over you, and leave with a printout of bone mineral content and density, fat mass and fat-free mass, with the density figures reported for defined regions such as the thigh bone and the spine [5].

The printout tells you most when you compare it with an earlier one from the same machine. The reading moves with the machine and its software [14], so the comparison worth making is the same person on the same scanner some months apart, and a body fat percentage from one clinic is not interchangeable with one from another. If the question is only whether body fat is high rather than exactly how high, a waist measurement fed into a published equation answered it better than body mass index did in 3,456 American adults [4].